Avoid using whitespace as an interpolation modifier

We currently support two number-sign modifiers in interpolation/format specs: `+` or ` ` ("If specified, prints this character in front of non-negative numbers"), same as C/C++ `printf`.

I think it's poor design for a few reasons, and thus I'd like it to be replaced with an underscore since that's visually similar enough.

- Visually, it looks wrong—since we feeble humans (and maybe LLMs also? lol) are used to processing whitespace-separated blocks of characters, an interpolation containing a space character would look visually jarring.
- I think it's reasonable to expect that external tooling attempting to parse RGBASM (in, yes, more limited ways, à la pret's `scan_includes`) would be doing the moral equivalent of `line.split()`, and possibly performing further processing on the resulting tokens (e.g. `splice()` on macro args or interpolations), and get tripped up by an interpolation straddling two such tokens. (...true, this argument does fall apart in the presence of strings.)

I acknowledge the rationale behind the choice of ` `: it's exactly the character that's being emitted, and there's precedent in that C accepts it as `printf` format specifiers. I think that `_` is visually similar enough, and I do not think that C is a good reference to follow as far as language design goes :P

Still, I would like to suggest deprecating ` ` in favour of `_`. I'm open to this being decided against if we end up preferring not to, though :)

Start by locating the interpolation/format-spec handling and any tests for the existing `+` and whitespace modifiers. Determine whether whitespace should be deprecated in favor of `_`, including the compatibility behavior and any diagnostics. Done means the project has a decided modifier policy and matching implementation and tests.

Written by the indexing model from the issue text.